alt Hacker News> key question is not just force balance: it’s whether the inextensible
This is definitely not astra, taking a guess this is 5.6, perhaps not even Sol, which does not reflect the state of the frontier (what the research was about).
And yes, the paper is already outdated
> And yes, the paper is already outdated
The paper is about the fact that the benchmark’s evaluator is prone to egregious incorrect rejections of what answers that it should accept. A new model will not invalidate that issue.